> 2021/03/31 21:31, Oliver Kiddle <opk@zsh.org> wrote:
> 
> On 23 Mar, "Jun. T" wrote:
>> Does anyone know a way to prevent '--' from being split in 'partial word completion'?
> 
> It is possible if you check for a -- in $PREFIX and don't add the other
> matches when it is found. That might subvert approximate completion,
> however. .... (snip)

Thanks for the analysis.
It seems 'r:|[_-][^-]=* r:|=*' works _slightly_ better, although I don't
know why. Or does this have some bad side effects?

> 2021/03/31 21:31, Oliver Kiddle <opk@zsh.org> wrote:
> 
> Was this the only outstanding issue for the contributed _gcc update?
> What's the status otherwise?

I've been trying to find a reasonable default for library_path.
We need to add /usr/lib/x86_64-linux-gnu on Ubuntu (and Debian?) and
/usr/lib64 and on Fedora/CentOS. Are there any OSs that are known to
use directories other than /usr/{,local/}lib? In the attached _gcc
I use the following for OSTYPE==linux-gnu:

      tmp=$(_call_program library-paths $words[1] -print-multiarch)
      if [[ $tmp != '' && -d /usr/lib/$tmp ]]; then
	library_path+=( /usr/lib/$tmp )
      elif [[ -d /usr/lib64 ]]; then
	library_path+=( /usr/lib64 )
      fi

> It does add
> a commented out line that assigns args2 without the file glob. That
> serves no useful purpose, was probably debug that got left in and should
> be stripped.

OP (Jacob Gelbman) wanted to use the commend-out version, but I argued that
it was not a good idea in a response to him.

Only differences of the attached _gcc from the 2nd version by OP are
the above three points (except for format changes etc.).

I will push this soon unless any problems are found.